← All projects

Prototype · 2025

RAG Toolkit

A retrieval library for RAG: query expansion, HyDE and rank fusion.

AI / LLM

A small library for building retrieval in RAG systems that combines several techniques to improve relevance beyond plain vector search.

Problem

A single vector-database query often misses relevant passages when the user phrases a question differently from the document.

Solution

Expanding a query into several variants with de-duplication, hypothetical documents (HyDE), merging results with Reciprocal Rank Fusion, and splitting text into overlapping chunks.